Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N5/00—Computing arrangements using knowledge-based models
  • G06N5/04—Inference or reasoning models
  • G06N5/048—Fuzzy inferencing
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N20/00—Machine learning
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
  • G06F16/90—Details of database functions independent of the retrieved data types
  • G06F16/95—Retrieval from the web
  • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
  • G06N5/00—Computing arrangements using knowledge-based models
  • G06N5/02—Knowledge representation; Symbolic representation

Definitions

• the present application relates to computer implemented data processing techniques and, more specifically, to populating content into an online feed.
• Webpages posted by online publishers have a number of user interface elements that are used to distribute content to visitors of the webpage.
• One such user interface element is a newsfeed, also referred to as a feed.
• a feed comprises a seemingly never-ending sequence of content items or links to content items that are selectable by the user.
• online publishers are incented to have users view as many content items in the feed as possible. As such, online publishers populate the feed using a greedy algorithm that places content items having a higher click-through rate (CTR) near the top of the feed.
• CTR click-through rate
• Viewers may navigate away from the feed after only viewing a few content items, meaning that they have viewed fewer content items than the online publishers would like them to have viewed.
• FIG. 1 is an illustration of a feed, according to various embodiments.
• FIG. 2 is a block diagram of a scrolling behavior system, according to various embodiments.
• FIG. 3 is a flowchart of a method of generating and training a behavioral model for viewing feeds, according to various embodiments.
• FIG. 4 is a flowchart of a method of populating a feed based on the behavioral model, according to various embodiments.
• FIG. 5 is a block diagram that illustrates a computer system upon which an embodiment may be implemented.
• a feed displayed vertically is described herein.
• a height of the content item refers to the vertical height of the content item.
• the distance of the content item from the top of the feed is a vertical distance from the top of the feed.
• Other embodiments may include a feed displayed horizontally.
• the height of the content item refers to the horizontal width of the content item.
• the distance in a horizontally-displayed feed is a horizontal distance from the start of the feed, typically a rightmost edge of the feed. In some geographic locations or languages, the "top" of the horizontally displayed feed may refer to the leftmost edge of the feed.
• Online feeds are popular user interface elements because they allow user to scroll through and view content items efficiently without disruptions like reloading the webpage or navigating to a "next" webpage.
• a successful online feed is one where the user continues to scroll for more content items after viewing the first displayed content items in the feed.
• a behavioral modelling system is used to optimize the user's viewing of the entire feed rather than just the first few content items.
• Various characteristics of the content items can be used to place the content items such as click-through rate (CTR), the height (h) of the content item when displayed in the feed, the distance (t) from the top of the webpage to the top of the content item, and the size (s) of the screen on the user device when the item is shown.
• CTR click-through rate
• h height
• t the distance from the top of the webpage to the top of the content item
• size (s) of the screen on the user device when the item is shown a second metric
• CSP predicted continuous scrolling probability
• the CTR of the content item is zero if the content item is only partially shown or not shown on the screen.
• the content items are shown in the decreasing order of expected reward from the top of the feed downwards: A then B then C then D. Since the height of content item A (h_A) is lOOpx, which is equal to the size of the screen (s), content item A occupies the screen completely, and items B, C and D are all pushed lower in the feed, below the displayed area. This results in a total revenue of $100.
• FIG. 1 is an illustration of a part of a feed 100, according to various embodiments.
• the feed 100 is depicted as a vertical feed but may be a horizontal feed in some embodiments.
• the feed 100 comprises an optional header 102 and content items 104-112. At a given time, only a portion of the feed 100 can be displayed on a screen of a user device (not depicted).
• the feed 100 may be navigated by the user by scrolling down to view more content items in the feed 100 or scrolling up to view previously viewed content items in the feed 100.
• the feed 100 comprises an optional header 102 located at the top of the feed 100.
• the header 102 is not a content item itself.
• the header 102 may include a logo of, or otherwise identify, the webpage providing the feed 100 and a menu or link to a menu of options selectable by the user.
• the header 102 can include other functionalities such as a search bar or notifications.
• the header 102 may persist at the top of the screen as the user scrolls down through the feed 100.
• the header 102 may be hidden from view and be re-presented upon receiving an input from a user, such as an instruction to scroll up or to stop scrolling.
• the feed 100 comprises a sequence of content items, such as content items 104- 112.
• the content items 104-1 12 are presented as an ordered sequence, where the content items are vertically arranged according to an ordering as described herein.
• the content items can include one or more links to further content items.
• the position of each of the content items 104-1 12 is expressed in terms of its vertical size and its distance from the top of the first content item 104 in the feed 100. Other embodiments can express the distance from the top of the header 102. To illustrate, for content item 1 10, its height, labelled h 1 18, is measured from the top of the content item 1 10 to the bottom of the content item 110 and expressed as a number of pixels.
• the distance of content items 104-108, labelled 1 116 is measured from the top of content item 104 to the top of content item 1 10 and expressed as a number of pixels.
• the feed 100 may comprise more or fewer content items than those depicted in FIG. 1.
• the screen size can depend on a number of variables controlled by the client device, such as font size, zoom level, size of the physical display, and whether a mobile website or a desktop website is accessed. While the screen size, s, can vary, for the purposes of the present disclosure, it is assumed to be constant number of pixels during the display of a feed 100. In FIG. 1 , the screen size s 114 is depicted as two parallel lines. The screen size is shown at the top of the feed 100.
• the screen size can be conceptualized as a floating window that encompasses the same number of pixel but includes different content items. Further, the screen size s does not necessarily conform to the sizes of the content items. As such, only portions of content items at the top or bottom of the displayed feed may be shown. For example, content item 108, as depicted, is bisected by the bottom line of s 114. As such, only the portion of content item 108 above the line is displayed at the user device unless the user scrolls down.
• FIG. 2 is a block diagram of a scrolling behavior system 200, according to various embodiments.
• the scrolling behavior system 200 generates, trains, and implements a behavior model used to order content items in a feed 100.
• the scrolling behavior system 200 uses the behavior model to sequence content items in feeds rendered to users.
• the description of the scrolling behavior system 200 describes how modules may be implemented. Upon reading this specification, a person having ordinary skill in the art may use the description herein as pseudo code for implementing the modules.
• a labelling model 202 labels content items 104-112 in a feed 100 displayed at a user device.
• the content items 104-112 may each be labelled “viewed”, “clicked”, and “scrolled”.
• a content item in the feed 100 is labelled “viewed” when it is completely or partially displayed at the user device.
• a content item in the feed 100 is labelled “clicked” when it is clicked by the user of the user device.
• a content item can be labelled "viewed” and labelled “scrolled” or just labelled "viewed”.
• the labels assigned to each content item are later used to train a behavioral model.
• a model training module 204 trains the behavioral model(s) based on the labelled content items and one or more additional inputs, such as user attributes and content item attributes.
• Two behavioral models can be trained. The first behavioral model estimates or outputs the probability of a content item being clicked if it is viewed. The second behavioral model estimates or outputs the probability of an item scrolled if it is viewed.
• the models are trained by various machine learning algorithms with historical data of users' feed visit behavior, and/or other features.
• a user attributes database 206 stores one or more attributes of each user who may request a feed 100.
• the user attributes can be quantified values describing one or more typical behaviors or characteristics of the user. Behaviors of the user may include a number of visits to the webpage over a period of time, a number of logins over a period of time, a number of content items viewed in the feed in one or more previous visits, a number of content items clicked on in the feed in one or more previous visits, and a number of content items scrolled in one or more previous visits.
• Attributes of the user may include gender, employment status, current employer, job title, a number of connections that the user has in a social network, whether the user has recently changed jobs or positions, membership in one or more affinity groups associated with the webpage, participation in a discussion or activity, the user's locale, an amount of time elapsed since the account was created, age of the user, and when a user's profile was last updated.
• a content item attributes database 208 stores one or more attributes in association with one or more content items that can be populated into a feed.
• the attributes can include, for example, a height h 118 of the content items, a number of times that the content items was viewed or labelled as viewed, a number of times the content item was scrolled or labelled as scrolled, a distance 1 116 of the content item from a top of the feed each time the content item was viewed or scrolled.
• Other attributes that may be used include, but are not limited to, the type of content such as article, photo, document, or advertisement; and the origin of the content such as from another user, from a famous or influential member of the website, or from sponsor of the website.
• a scoring module 210 uses the behavioral model, scores each of multiple content items that are candidates for inclusion in a particular feed.
• the set of candidate content items may be identical for each of multiple users or may be determined for each user or for each of multiple user groups, such as female users over 25 years old.
• the scoring results in a quantitative reward value associated with the content items, their respective placement in the feed, and their respective heights.
• the scoring is used to order the content items in the feed such that each content item is positioned to optimize the probability of being viewed for all content items in a feed. Because the user views more content items, the user is more likely to click through at least one content item. Over time, the user may prefer the webpage over other webpages and visit the webpage, and the feed, more often.
• FIG. 3 is a flowchart of a method 300 of generating and training a behavioral model for viewing feeds, according to various embodiments.
• the method 300 can be used to generate a behavioral model from a known set of user characteristics and current scrolling behavior of the users.
• the method 300 is performed when a previously generated feed 100 is displayed at a user device and includes labelling content items as viewed and scrolled.
• the method 300 can be performed by the scrolling behavior system 200.
• one or more attributes of a user are determined.
• the attributes of the user are retrieved from the user attributes database 206.
• the user attributes can include information about the past scrolling behavior of the user and past click through behavior of the user.
• the user attributes can be specific to the user or can be based on a group of users to which the user is assigned or with which the user is affiliated.
• attributes of one or more content items are determined.
• the attributes of the content items are retrieved from the user attributes database 208.
• the attributes can include a height h 118 of the content item, a history of distances 1 116 at which the content item has been positioned, and, for each distance 1 116, the screen size s 114 of the user device, and whether the content item was clicked.
• another attribute of the content item is the width or relative width of the content item in the feed 100 where the width of the content item is measured perpendicular to the height h 118 of the content item.
• Some embodiments may include a time spent by a user in viewing the content item in the feed.
• the time spent by the viewer can be automatically modified based on a length of the content item.
• the length of the content item can be measured according to the contents of a content item.
• the length of the content item is based on a playback time of the video.
• the length of the content item is based on a size of the image or a number of words in the image or a number of frames or slides associated with the image.
• a text content item e.g., an article or blog entry
• the length of the content item is based on a number of words in the content item.
• Various embodiments may include one or more complexity measures to determine the attributes of a content item.
• the complexity measure can include an average amount of time spent by a particular user viewing a plurality of content items before scrolling.
• Another complexity measure can include an average time spent viewing a particular content item by a plurality of users who viewed the particular content item.
• Another complexity measure identifies a number of images or videos included in, or associated with, a particular content item.
• the labelling module 202 labels the content items displayed on the user device as viewed.
• the content items are displayed depending, in part, on the distance 1 116 of the content item relative to the top of the feed 100.
• a larger screen size means that a greater number of the content items in the feed 100 may be concurrently displayed at the user device while a smaller screen size means that fewer content items may be concurrently displayed at the user device.
• the labelling module 202 labels each content item in the feed 100 as scrolled if one or more subsequent content items after the content item in the feed 100 is displayed at the user device.
• the labelling module 202 may label content items in the feed 100 as scrolled if a specified number of content item(s) immediately beneath the respective content item in the feed 100 is displayed at the user device. For example, a content item may be labelled scrolled if a third, fourth, fifth, sixth, seventh, or subsequent content item is displayed. Only content items labelled viewed can also be labelled scrolled.
• the content items labeled scrolled can be favored by the behavioral model as they are more likely to motivate the user to continue scrolling through a subsequent feed 100 and viewing more content items.
• the behavioral model is trained based the attributes and labels obtained in the method 300.
• the behavioral model models the probability of events that a content item is seen and/or clicked as a function of (i, u, h, t, s) where i is the content item, u is the user, h is the height of the content item, t is the distance from the top of the feed to the content item, and s is the size of the screen.
• the behavioral model optionally assumes that the higher (i.e., closer to top of the feed) a content item is placed, the higher the click-through rate (CTR) will be.
• CTR click-through rate
• the behavioral model optionally assumes that a content item has a higher chance of being displayed at the user device and/or clicked when it is completely shown in the first screen (i.e., the user doesn't scroll down in order to view the content item) than when it is not in the first screen (i.e., the user scrolls down to view the content item) because it requires less effort from the user:
• a user always views the content item at position (i-1) before viewing the content item at position i
• item i is seen] p_i is constant, independent of Pr[item i is seen] and its position.
• FIG. 4 is a flowchart of a method 400 of populating a feed 100 based on the behavioral model, according to various embodiments.
• the feed 100 is populated whenever the feed is rendered for the user.
• the feed may be rendered each time the user logs in to a particular webpage or each time the user navigates to the particular webpage from another webpage.
• one or more attributes of the user for whom the feed 100 is rendered are determined.
• the attributes of the user are retrieved from the user attributes database 206.
• the user attributes can include information about the past scrolling behavior of the user and past click through behavior of the user.
• the user attributes can be specific to the user or can be based on a group of users that the user is assigned to or affiliated with.
• attributes of one or more content items available for placement in the feed 100 are determined.
• the attributes of the content items are retrieved from the content attributes database 208.
• the attributes can include a height h 1 18 of the content item, a history of distances 1 116 at which the content item has been positioned, and, for each distance 1 1 16, the screen size s 114 of the user device, and whether the content item was clicked or otherwise selected.
• the scoring module 210 uses the behavioral model of method 300 to generate a score for each available content item where all of the attributes of the one or more content items are included in the scoring.
• the behavioral model can score the content items using only the scrolled labels of the operation 308.
• the behavioral model can score the content items using the scrolled labels of the operation 308 and the viewed labels of operation 306.
• the behavioral model can score the content items using the scrolled labels of the operation 308, the viewed labels of operation 306, and an established click-through rate associated with the content items.
• the scoring module 210 formulates the following optimization problem with the objective function being the expected reward of the whole visit:
• the scoring module 210 is optimizing over all possible permutations of the candidate items. Because of the product form of F_i, the scoring module 210 may be used to show that the optimal solution is to rank items based on
• (l-p_i) e A ⁇ -a h_i ⁇ is interpreted as the probability the user does not click on item i and scroll down to item i+1, under condition that item i has been seen. So the denominator is the probability that either the user clicks on item i or does not click and scrolls down to the next item. In other words, the denominator is the conditional probability that, given item i has been seen, the user does not see item i+1 and the visit terminates at item i. Therefore, p i/ ( 1 - (l-p_i) e A ⁇ -a h_i ⁇ ) is the CTR of item i, given that a visit terminates at item i.
• a more generalized optimization formulation and proof of optimal ordering of content items in the feed 100 is used.
• xj be the feature vector for item j (j being item ID), including display size features h and t.
• y_i be the feature vector of the item placed at the ith position (i being rank in feed).
• notations like x i.h to y i.h refer to the particular feature (size in this case) for item i and the item placed at position i.
• Y_i ⁇ y_l , y_2, y_i ⁇ denotes all the features for items placed up to the ith position.
• F i(Y i) is defined as the probability of the item at position i being clicked given all features of items displayed at position 1 to i.
• the scoring module 210 further assumes F_i(Y_i) follows a product form:
• g(y_i) Pr[ item at position i+l is seen
• the scoring module 210 relies on the following assumptions that:
• the scoring module 210 can formulate the following optimization problem to maximize the expected reward of the whole visit:
• Wi Xi .r
• the display module 212 uses the solution determined by the scoring module 310, the display module 212 generates the feed 100 based on the relative scores of each of the content items.
• the content items are laid out from highest scoring at the top of the feed 100 with successively lower-scoring content items laid out below.
• the feed 100 is rendered at the user device.
• the techniques described herein are implemented by one or more special-purpose computing devices.
• the special-purpose computing devices may be hard-wired to perform the techniques, or may include digital electronic devices such as one or more application-specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) that are persistently programmed to perform the techniques, or may include one or more general purpose hardware processors programmed to perform the techniques pursuant to program instructions in firmware, memory, other storage, or a combination.
• ASICs application-specific integrated circuits
• FPGAs field programmable gate arrays
• Such special-purpose computing devices may also combine custom hard-wired logic, ASICs, or FPGAs with custom programming to accomplish the techniques.
• the special-purpose computing devices may be desktop computer systems, portable computer systems, handheld devices, networking devices or any other device that incorporates hard-wired and/or program logic to implement the techniques.
• FIG. 5 is a block diagram that illustrates a computer system 500 upon which an embodiment may be implemented.
• Computer system 500 includes a bus 502 or other communication mechanism for communicating information, and a hardware processor 504 coupled with bus 502 for processing information.
• Hardware processor 504 may be, for example, a general purpose microprocessor.
• Computer system 500 also includes a main memory 506, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 502 for storing information and instructions to be executed by processor 504.
• Main memory 506 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor 504.
• Such instructions when stored in non-transitory storage media accessible to processor 504, render computer system 500 into a special-purpose machine that is customized to perform the operations specified in the instructions.
• Computer system 500 further includes a read only memory (ROM) 508 or other static storage device coupled to bus 502 for storing static information and instructions for processor 504.
• ROM read only memory
• a storage device 510 such as a magnetic disk, optical disk, or solid-state drive is provided and coupled to bus 502 for storing information and instructions.
• Computer system 500 may be coupled via bus 502 to a display 512, such as a cathode ray tube (CRT), for displaying information to a computer user.
• a display 512 such as a cathode ray tube (CRT)
• An input device 514 is coupled to bus 502 for communicating information and command selections to processor 504.
• cursor control 516 is Another type of user input device
• cursor control 516 such as a mouse, a trackball, or cursor direction keys for communicating direction information and command selections to processor 504 and for controlling cursor movement on display 512.
• This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), that allows the device to specify positions in a plane.
• Computer system 500 may implement the techniques described herein using customized hard-wired logic, one or more ASICs or FPGAs, firmware and/or program logic which in combination with the computer system causes or programs computer system 500 to be a special-purpose machine. According to one embodiment, the techniques herein are performed by computer system 500 in response to processor 504 executing one or more sequences of one or more instructions contained in main memory 506. Such instructions may be read into main memory 506 from another storage medium, such as storage device 510. Execution of the sequences of instructions contained in main memory 506 causes processor 504 to perform the process steps described herein. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions.
• Non-volatile media includes, for example, optical disks, magnetic disks, or solid-state drives, such as storage device 510.
• Volatile media includes dynamic memory, such as main memory 506.
• Storage media includes, for example, a floppy disk, a flexible disk, hard disk, solid- state drive, magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with patterns of holes, a RAM, a PROM, and EPROM, a FLASH-EPROM, NVRAM, any other memory chip or cartridge.
• Storage media is distinct from but may be used in conjunction with transmission media.
• Transmission media participates in transferring information between storage media.
• transmission media includes coaxial cables, copper wire and fiber optics, including the wires that comprise bus 502.
• Transmission media can also take the form of acoustic or light waves, such as those generated during radio-wave and infra-red data communications.
• Various forms of media may be involved in carrying one or more sequences of one or more instructions to processor 504 for execution.
• the instructions may initially be carried on a magnetic disk or solid-state drive of a remote computer.
• the remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem.
• a modem local to computer system 500 can receive the data on the telephone line and use an infra-red transmitter to convert the data to an infra-red signal.
• An infra-red detector can receive the data carried in the infra-red signal and appropriate circuitry can place the data on bus 502.
• Bus 502 carries the data to main memory 506, from which processor 504 retrieves and executes the instructions.
• the instructions received by main memory 506 may optionally be stored on storage device 510 either before or after execution by processor 504.
• Computer system 500 also includes a communication interface 518 coupled to bus 502.
• Communication interface 518 provides a two-way data communication coupling to a network link 520 that is connected to a local network 522.
• communication interface 518 may be an integrated services digital network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection to a corresponding type of telephone line.
• ISDN integrated services digital network
• communication interface 518 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN.
• LAN local area network
• Wireless links may also be implemented.
• communication interface 518 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.
• Network link 520 typically provides data communication through one or more networks to other data devices.
• network link 520 may provide a connection through local network 522 to a host computer 524 or to data equipment operated by an Internet Service Provider (ISP) 526.
• ISP 526 in turn provides data communication services through the world wide packet data communication network now commonly referred to as the "Internet" 528.
• Internet 528 uses electrical, electromagnetic or optical signals that carry digital data streams.
• the signals through the various networks and the signals on network link 520 and through communication interface 518, which carry the digital data to and from computer system 500, are example forms of transmission media.
• Computer system 500 can send messages and receive data, including program code, through the network(s), network link 520 and communication interface 518.
• a server 530 might transmit a requested code for an application program through Internet 528, ISP 526, local network 522 and communication interface 518.
• the received code may be executed by processor 504 as it is received, and/or stored in storage device 510, or other non-volatile storage for later execution.

Priority Applications (1)

Applications Claiming Priority (2)

Publications (1)

Family

ID=57882590

Family Applications (1)

Country Status (3)

Families Citing this family (4)

Citations (5)

Family Cites Families (6)

• 2015
  • 2015-07-30 US US14/814,104 patent/US10275716B2/en active Active
• 2016
  • 2016-07-22 CN CN201680038910.6A patent/CN108604232B/zh active Active
  • 2016-07-22 WO PCT/US2016/043698 patent/WO2017019548A1/en active Application Filing

Patent Citations (5)

Also Published As

Similar Documents

Legal Events